“On a current podcast episode, you may hear Joe Rogan ask OpenAI founder Sam Altman whether or not society must be involved about pretend AI content material. “How will we be capable to inform what’s actual and what’s not?” It was a very becoming query for the episode — because the complete hourlong dialog was generated by AI.”
Relying on whom you ask, AI is both an existential menace to each artistic trade, a groundbreaking expertise that can advance human civilization, or a form of enjoyable, dumb toy that’s in the end a pale imitation of the actual deal. The creator of The Joe Rogan AI Expertise, a form of AI-generated fanfic YouTube podcast that depicts imaginary conversations between Rogan and a brand new visitor each episode, form of believes all three. Or at the least, he brings up all three over the course of a virtually hourlong dialogue that Scorching Pod had with him earlier this month.
Hugo, who requested us to solely use his first identify to guard his id, is a artistic director for a VFX studio and advert company in Sydney, Australia. Hugo is fast to confess that his YouTube podcast is only for enjoyable and never meant as any actual menace to the actual Rogan or precise flesh and blood podcasters.
“That is purely fiction — only for enjoyable — so don’t go round making your little TikToks and make the world consider I mentioned issues I by no means mentioned,” the AI clone of Joe Rogan AI warns within the first episode of the podcast, which Hugo revealed a month in the past. In it, the pretend Rogan and an equally pretend Altman have a ChatGPT-scripted dialog. The ensuing podcast was eerie sufficient that Rogan addressed it himself.
“That is going to get very slippery, children,” Rogan — the actual one — wrote in a tweet.
The concept for The Joe Rogan AI Expertise got here to Hugo again in February, when social media discourse round ChatGPT appeared to be at an all-time excessive. He had been commonly following AI developments over the previous yr and watched “a few YouTube movies about voice cloning” when the thought popped into his head.
“What if I attempted to generate an entire The Joe Rogan Expertise podcast with ChatGPT after which managed to make use of a type of voice cloning platforms?” he thought.
After some Google looking, he discovered various AI voice software program corporations with demo samples of Joe Rogan’s voice. Sadly, quite a lot of them weren’t that nice or obtainable for public use. Lastly, after a few weeks of looking, Hugo discovered a text-to-speech platform that may clone any voice primarily based on snippets of audio. He used that platform for the voice of Rogan and different visitors on the podcast, which have additionally included Donald Trump and Andrew Tate.
Supporters of AI say it is going to be quicker and cheaper to supply than human-made work. However the work of prompting a usable script on ChatGPT, throwing it right into a text-to-speech program, and enhancing audio — all takes hours of human work.
“The primary […episode] took me every week, the second took me a day and a half, and the final one took me greater than three weeks,” wrote Hugo in an e-mail. “It’s very depending on many various points however the principle one is my very own important opinion of the episode. If it’s not adequate in my eyes, I’ll spend quite a lot of time making it proper.”
Hugo spent a very long time gathering pattern audio and perfecting his AI-generated voices. Whereas creating the primary voice clone of Rogan was straightforward, Hugo spent hours refining it to perfection in later episodes. Even then, there can be moments when his audio system sounded robotic or AI-like. Enhancing the script to create one thing that seems like a standard human dialog additionally took work.
“[…ChatGPT] is de facto not that unbelievable but, actually, with replicating the speech patterns of somebody,” mentioned Hugo. Regardless that somebody like Rogan has numerous hours of voice recordings obtainable on the web — ChatGPT solely trains on written info. Even feeding it transcripts of interviews, in keeping with Hugo, isn’t foolproof.
“While you actually take heed to a dialog between two folks, you’ll see that they discuss over one another loads. They reply one another — with very transient solutions typically — and it’s only a fixed backwards and forwards,” he mentioned.
Making an AI-generated podcast of Joe Rogan little doubt takes Hugo longer than the precise Joe Rogan simply doing it himself. So why make an AI-generated podcast of one of the crucial common podcasters in any respect?
“I needed to create a model of the Joe Rogan podcast that will by no means occur, or hasn’t occurred but,” Hugo mentioned. Which is true — Rogan has but to interview Altman or any of the opposite visitors which are depicted in The AI Expertise. Hugo believes it’s the present’s most important draw — to stage hypothetical conversations between Rogan and different folks. He ultimately desires to attempt creating episodes that depict lifeless visitors, similar to Steve Jobs or Abraham Lincoln.
I spent an hour or so listening to segments of The Joe Rogan AI Expertise. Whereas the standard of the voice cloning is uncanny and Hugo’s enhancing is expert-level, the precise content material leaves loads to be desired. The dialogue is wordy and bland in a manner we now affiliate with AI-generated textual content.
“To be trustworthy personally, I feel we’re not wherever close to near changing actual podcasts or actual dialog but, however I can perceive the worry of individuals. As a result of I don’t assume it’s tremendous shut — I imply — it’s shut. However in the event you’ve watched quite a lot of Joe Rogan’s podcast, you may really feel right away that […The AI Experience] shouldn’t be actual and lacks quite a lot of emotion and proper intonations,” mentioned Hugo.
The primary episode of The Joe Rogan AI Expertise wracked up near half one million views (seemingly as a consequence of Rogan tweeting it), however viewership has dropped in subsequent episodes. That’s in distinction to AI remixes of songs by Drake and different artists, which have shortly gone viral, however the jury remains to be out on whether or not people will truly want AI recreations to the actual factor.
There are different causes to be involved concerning the rise of AI in podcasting. Digital fakes might make Rogan or another host say one thing objectionable. And Hugo says he’s already seeing how AI instruments might change sure jobs within the VFX trade.
However that doesn’t imply they spell the tip of live-hosted exhibits. At one level, I introduced up a Vulture interview with Comedy Bang! Bang! host Scott Aukerman, the place he notes that podcasts stand out as a result of a lot of leisure right now is prescripted. Podcasts are uncommon in you can hear folks giggle and “be joyful within the second.”
Hugo doesn’t consider AI can change that spontaneity.
“You realize, the factor whenever you’re with a pal at evening and also you’re speaking, and an hour later you find yourself in a totally totally different place from the place you began? And also you notice AI can’t actually try this. The way in which the human mind works — it’s very unpredictable.”
